Issue link: https://iconnect007.uberflip.com/i/1183414
NOVEMBER 2019 I DESIGN007 MAGAZINE 21 and cost constraints. The "rules of thumb" that work in the server world often won't work here, and I think this is an interesting opportu- nity for EDA tools. EDA tools have typically fo- cused on design performance, not cost. What would EDA tools for a cost-sensitive design look like? I'm not sure I know, but I'm work- ing on finding out. Shaughnessy: What advice would you give a new designer or an EE just entering this seg- ment? Westerhoff: Try to take the time to understand how things work and build up your engineer- ing intuition. Today's designers are surround- ed by sophisticated simulation tools, but those tools are a poor substitute for human under- standing and insight. Besides, if you don't have a strong grasp of the physics and the be- havior you expect, how do you know if your simulation results are believable or not? Sim- ulators are great, but they're still just tools. I don't expect a simulator to make a design deci- sion any more than I expect a hammer will be able to drive nails by itself. Designers who can articulate the physics behind why something behaves the way it does—and combine that knowledge with practical advice of what de- sign tradeoffs to make and how to verify them through analysis—will be highly valuable in any environment. Shaughnessy: Is there anything you'd like to add? Westerhoff: I have a great line from an old com- puter manual to share. By way of background, I'm a bit of a computer historian, fascinated by how computers evolved in the late 1950s and early 1960s. The IBM 650 RAMAC was IBM's first commercial computer introduced in 1954 with a "bi-quinary" architecture. In those days, computer manufacturers were still trying to ex- plain to the public what computers were, in contrast to the "giant-brain" images that sci- ence fiction writers and the media had por- trayed. One of the programming manuals for the 650 water-cooled machine with 128 GB of RAM, an ultra-fast M.2 NVMe and 850W power supply, a modern graphics card, case, keyboard, etc., for about $3,000. I learned some hard lessons about practical (as opposed to theoretical) memory bandwidth and how it drives compute throughput—things I've been exposed to for years but never fully understood. I ended up with a machine that does a decent job of running 3D EM simula- tions, which was the original goal, after all. Shaughnessy: How do you think designers, and perhaps EDA tool companies, will be affected by new technologies like 5G and IoT? Westerhoff: I think that we're living in an in- creasingly client/server world, where electron- ics technology is embedded in pretty much everything. That drives design in two differ- ent directions, even though they're usually two sides of the same coin. On the server-side, designs are driven by per- formance, capacity, and reliability; cost is usu- ally secondary. These are the design applica- tions that EDA has typically targeted, and these are the applications with which SI/PI/EMC ex- perts are usually associated. Each major new application—112G, 5G, take your pick—brings new design issues that the industry has to work through and bring to production. The de- sign challenges keep getting tougher, but the broader pattern has been repeating since the dawn of EDA. The client-side is more or less the opposite, even though the core technology is the same, with 5G, let's say. This is typically the consum- er side of the equation; cost and time to market are the prime drivers. The product has to work, but that's it. Providing 10% more range in a cellphone won't substantially increase sales or justify a 10% increase in the phone's price, for example. These are also the applications where the designers may not have enough support from the SI/PI/EMC experts because they're busy on the other side of the fence. So, how do these products get designed? De- signers still need to make tradeoffs, and those tradeoffs are often more extreme due to space